For the better part of two years, U.S. grid operators have been staring at the same alarming projection: artificial intelligence data centers would devour so much electricity, so fast, that summer blackouts could follow. Federal regulators flagged the risk publicly. Utilities scrambled to keep aging power plants running. State commissions debated emergency procurement.
Now, heading into the summer of 2026, the crisis looks different than expected. The massive data center buildout that was supposed to strain the grid by this point has fallen behind schedule, slowed by construction delays, transformer shortages, permitting logjams, and financing complications. Many of the facilities that forecasters assumed would be drawing power by now exist only on paper. The result: grid operators have more breathing room than they anticipated, and the immediate threat of rolling blackouts has softened for tens of millions of households.
The demand is not gone. It is late. And that distinction matters enormously for what comes next.
Federal regulators sounded the alarm early
The Federal Energy Regulatory Commission laid out the concern in its 2025 Summer Assessment, published last year. FERC staff identified tightening reserve margins across several regions and pointed squarely at the source. “Load increases are largely due to hyperscale users, such as data centers,” the agency stated. The message was direct: electricity demand was growing faster than new generation or transmission could keep up, and the single largest category of new load came from facilities built to train and run AI models.
That warning shaped planning decisions across the industry. Regional transmission organizations adjusted capacity procurement targets. State regulators questioned whether coal and gas plants scheduled for retirement should stay online longer. The working assumption was that data center operators would energize new campuses roughly on their stated timelines, stacking gigawatts of constant, around-the-clock demand onto grids that already ran thin during heat waves.
FERC carried the same concern forward into its 2026 Summer Assessment and accompanying staff presentation. The continuity between the two reports signals that federal analysts view this not as a one-off spike but as a structural shift in how the country consumes power.
The demand wave stalled
Separately, the U.S. Energy Information Administration published scenario-based analysis examining how different rates of data center growth would ripple through fuel markets, generation dispatch, and wholesale electricity prices. Under faster-growth assumptions, fossil fuel generation would climb and natural gas prices would rise. Under slower-growth assumptions, the pressure would ease considerably, reducing the need to fire up older, less efficient plants during peak hours.
The slower path now looks closer to what is actually happening. Interconnection timelines for large data center projects have stretched well beyond the schedules utilities plugged into their demand forecasts. Several forces are responsible:
- Transformer shortages. Large power transformers, the critical link between high-voltage transmission lines and a data center’s internal electrical systems, carry lead times that the U.S. Department of Energy has flagged as a national vulnerability. Orders placed in 2023 and 2024 are, in many cases, still awaiting delivery.
- Permitting and interconnection backlogs. PJM Interconnection, the grid operator covering 13 states from Virginia to Illinois, has publicly acknowledged a queue of more than 2,500 projects totaling hundreds of gigawatts. Most will never be built, but the sheer volume of applications has slowed the engineering studies required before any single project can proceed.
- Construction and financing friction. Rising interest rates through 2024 and early 2025 increased borrowing costs for capital-intensive builds. Site preparation, water access for cooling, and local permitting have added months to timelines that tech companies initially presented as aggressive but achievable.
In practice, many of the megawatts regulators expected to see on the grid by the summers of 2025 and early 2026 are still in development rather than drawing power.
Where the pressure is sharpest
National averages obscure the places where data center demand is most concentrated. Northern Virginia’s “Data Center Alley” in Loudoun and Prince William counties remains the densest cluster of hyperscale facilities on the planet, and Dominion Energy has repeatedly warned that local transmission capacity is under strain. Central Texas, where ERCOT operates an isolated grid with no ability to import power from neighboring regions during emergencies, faces its own version of the problem. The Phoenix metro area and parts of central Ohio are also absorbing large volumes of planned data center construction.
FERC’s assessments address reliability at the level of NERC assessment areas, broad geographic zones that can mask localized stress points. A single large campus coming online in a constrained corridor can materially shift the supply-demand balance for surrounding communities, even if the regional picture looks comfortable. Local substation capacity, distribution-line upgrades, and the availability of nearby generation all determine whether a given neighborhood feels the squeeze.
The delay is not a cancellation
Tech companies have not pulled back from their expansion plans. Microsoft, Google, Amazon, and Meta have each committed tens of billions of dollars to AI infrastructure through the end of the decade. Many of those investments still depend on large, grid-connected campuses. If permitting accelerates, transformer deliveries catch up, or financing conditions improve, the demand that failed to materialize in 2025 and early 2026 could arrive in a compressed window.
That scenario worries some grid planners more than a gradual ramp would. Generation and transmission projects designed to meet earlier demand timelines may themselves have slipped, meaning the supply side could be just as late as the demand side. A sudden burst of data center energizations, concentrated in a few regions over a short period, would be harder to absorb than the steady climb utilities originally modeled.
Some companies are also exploring ways to sidestep the grid entirely. On-site natural gas turbines, long-duration battery storage, and early-stage investments in small modular nuclear reactors all reflect an industry hedging against the possibility that grid connections will remain slow and unreliable. If a meaningful share of future data center load is met behind the meter, the public grid would see less demand growth than current forecasts assume, but it would also collect less revenue to fund the transmission upgrades those same forecasts call for.
What consumers should watch
Whether the current breathing room translates into lower electricity bills is far from certain. Wholesale power markets respond to expected future demand, not just present consumption. If traders and generators still anticipate a near-term data center buildout, forward prices may stay elevated even as real-time conditions ease. The EIA’s scenario analysis showed that generation costs and natural gas prices are sensitive to the pace of data center growth, but the agency did not publish a timeline for when slower uptake would filter through to retail rates.
State-level rate cases, utility fuel-cost adjustments, and long-term power purchase agreements all sit between wholesale market shifts and the number on a household electric bill. Consumers in states with vertically integrated utilities may see changes on a different schedule than those in deregulated markets. And in regions where utilities have already invested in grid upgrades to accommodate data centers that have not yet arrived, ratepayers could end up covering the cost of infrastructure that sits underused for years.
Why the grid’s reprieve may be shorter than it looks
The public record does not yet contain a comprehensive, independently audited accounting of how many megawatts of data center load have actually come online versus how many were expected by mid-2026. That gap limits how confidently anyone can declare the AI power crisis over. What the federal data does support is a narrower finding: the immediate reliability risk that regulators feared heading into the mid-2020s cooling seasons has eased. Reserve margins look healthier than worst-case projections suggested. Grid operators have more time to prepare.
But the underlying forces driving long-term electricity demand, including billions of dollars in committed capital spending, a global race to scale AI capabilities, and a generation fleet that is retiring faster than it is being replaced, remain firmly in place. The surprise break is real. How long it lasts depends on whether the delays that created it are temporary friction or a sign that the buildout was always going to be slower and messier than the industry promised.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
